01
Motion in
Data from any wearable, phone, or self-report. Six device formats supported, each built against a real file from that device.
Tremora
Verified numbers built for clinical review.
Tremora turns movement data into a decision-ready read for the neurologist, and shows exactly where every value came from.
Why Tremora Exists
Four FDA-cleared products already measure tremor from a wearable.
Eight percent of Parkinson's patients use any of them. Of those, three in ten say it changed how their care team managed them.
An entire category solved measurement, and almost nobody's treatment is different.
- 8 in 100
- Use any medical wearable
- 3 in 10 of those
- Say it changed their care
How It Works
02
Tremora scores it
Isolates true tremor from ordinary movement and scores it on one index.
03
Every number carries its source
Measured, calculated, inferred, or patient-reported. Attached at ingestion. A value without a source can't display at all.
04
Decision-ready read
Trends, medication response, and a plain-language summary. When the data can't support a value, it says so.
The Same Number, Twice
Same number. Opposite decision.
The same value, 2.4, shown two ways. Measured, four hours of continuous wear straight off the sensor: act on it. Inferred, from two days the patient barely wore the device: wait.
2.4
Measured
4 hours continuous wear
Act on it
2.4
Inferred
2 days, barely worn
Wait
No other tool can tell a clinician which one they are looking at.
Where did this number come from?
Every displayed value can be expanded into the record it was derived from.
- Sessions contributing
- 3 sessions, Jun 13 to Jun 15, 4h 12m of qualifying wear
- Windows excluded
- 148 windows dropped: 96 below the wear-rate floor, 52 flagged as non-tremor motion
- Quality verdicts
- Sample-rate check passed, gap check passed, saturation check passed
- Calibration state
- Amplitude units calibrated for this sensor
- Provenance
- Measured — exported by the module that derived the value
This is the product's answer to why a number is what it is.
The Problem
Less than 1 hour of tremor data is collected per year for the average patient.
Tremor changes hour to hour with medication and fatigue. A clinic visit captures a few minutes of it, a few times a year, and misses the rest. Between visits, the neurologist is working blind.
~0M
Americans living with Parkinson's disease
0.00%
Pooled prevalence of essential tremor across all ages, rising to 5.79% at 65 and over
Clinic visit
~10 minutes per quarter
Tremora
continuous, every day
What Tremora Does
How Tremora turns raw movement into a full clinical picture.
Tremora estimates tremor from motion data, tracks severity on the Tremora Severity Index, aligns symptoms to dose events, and produces a summary of the logged interval. Parkinson's and essential tremor share one record schema.
All Features
What you get on the Tremora dashboard.
Answer-first patient page
Opens with interval change: what moved, prior value, current value, and window.
Medication-response tracking
Dose events aligned to tremor severity across the logged day.
Severity over time
Week-level trend on the Tremora Severity Index.
Symptom timeline
Within-day and between-visit timing of higher-severity windows.
Phenotype hint
Rest / postural / kinetic classification as an estimate, not a diagnosis.
Long-term history
Monthly means with gaps left blank; captions state values only.
Provenance on every number
Measured, estimated, or patient-reported; mark is required for display.
Ask Tremora
Record-grounded Q&A. No diagnosis, prediction, or treatment directives.
Works with or without hardware
Sense, phone IMU, or self-report. Absent channels are not shown as zero.
Appointment-ready notes
Structured clinical note generated from each ingested session.
Visit prep sheet
One-page printable summary; provenance marks survive print.
Cohort view
Panel list with a needs-attention surface; empty when none apply.
Medication Curve
Visualizing dosage time's impact on tremor severity.
Common tremor medications have onset, peak, and wear-off phases. Tremora overlays severity against logged dose times so response timing is visible in the record.
Universal Compatibility
One ingestion path. Any motion source.
Tremora is not locked to one vendor's hardware. Movement data flows through a single pipeline that normalizes it into one record, regardless of where it came from, and every value carries a provenance mark showing its origin. A clinical wearable, a consumer watch, a phone, or self-report all enter the same way.
Designed to accept motion data from sources like:
- Apple Watch
- PKG Watch
- Fitbit
- STAT-ON
- Garmin
- Kinesia 360
- Samsung Galaxy Watch
- ActiGraph
- Google Pixel Watch
- APDM Opal
- Empatica
- Shimmer
- Withings
- Axivity
- WHOOP
- Physilog
- Oura
- Byteflies
- Smartphone IMU
- MC10 BioStamp
- Tremora Sense
Refusal Behavior
When the data can't support a value, the system says so.
Tremora detects six data-quality failure modes and refuses to produce a value below an evidence floor. A missing number is not an outage; it's the honest answer, and it's the one thing a clinician can't get from a tool that always returns something.
Mon
2.1
Tue
2.4
Wed
Not enough data
Thu
2.2
Fri
1.9
Sat
Rate too low
Sun
2.6
Mon
2.5
Tue
Not enough data
Wed
2.3
Thu
2.0
Fri
Not enough data
Sat
2.2
Sun
2.4
Detected and gated before anything is scored
- Sample rate below threshold
- Wear time below the evidence floor
- Gaps in the recorded stream
- Sensor saturation
- Non-tremor motion dominating the window
- Uncalibrated amplitude units
Methods
Built on methods we can show.
Our tremor scoring is evaluated on public movement datasets with leakage-controlled validation, and our methods are written up and reproducible. We show our work instead of quoting a single number.
Clinical pilots beginning with partner sites.
Our Transparency
Built to be trusted. Every number shows its source.
Displayed values carry a provenance mark. Generative output is limited to summarizing and answering questions about the record. Diagnosis, prediction, and treatment directives are blocked before display.
80 of 80. Unsafe outputs blocked before display.
Unsafe outputs blocked before display.
We wrote 46 adversarial prompts designed to make the AI layer diagnose or recommend treatment. It blocked zero of them. We rebuilt it and it now blocks all 46. A second pass added 34 more attacks aimed at meaning rather than wording; 22 got through before the fix. It now blocks all 80, with zero false positives on 26 controls, and fails closed on anything ambiguous.
Zero guesswork with numbers
Each value is labeled by how it was obtained. Line style encodes the same distinction, so the further a value sits from direct measurement, the more broken its line.
Every value is labeled by how it was obtained.
- Measured2px solid — read off a sensor
- Estimate1px solid — calculated by our model
- Proxy1.5px dashed — inferred from a related signal
- Self-report1.5px dotted — the patient told us
The more removed a value is from direct measurement, the more broken its line.
Resting tremor, morning
2.4 TSI
Daily OFF-state estimate
3.1 h
Patient-reported dyskinesia
mild
Line style carries the meaning: solid, dashed, or dotted.
Defining chatbot boundaries.
The model may write clinical summaries and answer record-grounded questions. A deterministic gate rejects diagnose / predict / treat language. Ambiguous cases fail closed.
The assistant can summarize the record. It can't diagnose or direct care. That boundary is enforced in code, not by prompt.
- Allowed
Record summary
Summarize the last two weeks.
- Blocked
Diagnosis language
Does this patient have Parkinson's?
- Blocked
Treatment directive
Increase the evening dose.
- Ambiguous, blocked
Ambiguous request
Is the medication still working?
80 adversarial inputs · 80 blocked · fails closed
Allowed to summarize. Not allowed to diagnose or direct care.
Why This Matters Right Now
A huge population, and almost no one serving it.
Only 9% of Medicare Parkinson's patients see a movement disorder specialist. Half are managed by a general neurologist. There are roughly 660 specialists in the entire country. Most patients are therefore followed with no objective record of what happens between visits. That's the gap Tremora fills.
Who actually manages these patients
Share of Medicare Parkinson's patients seen by a movement disorder specialist. Roughly 660 such specialists practise in the entire country.
9%
See a specialist
9%
Seen by a movement disorder specialist
9%
91%
Managed elsewhere, about half in general neurology
91%
Where existing tools stop short
Tools clinics encounter often share one or more of these constraints.
Device-locked
Tied to one wristband or one vendor's hardware.
Condition-locked
Built for a single diagnosis, not both PD and ET.
Research-only
Made for labs, not an ordinary clinic workflow.
See our dashboard in action.
Check out our live demo with no login required. Test out Tremora's features using sample patients running on the real backend.